Improving biomedical named entity recognition with syntactic information.

Area of Science:

• Computational biology
• Natural Language Processing
• Bioinformatics

Background:

• Biomedical Named Entity Recognition (BioNER) is crucial for text comprehension but faces challenges due to limited labeled data and domain expertise.
• Existing BioNER models often struggle to flexibly incorporate external knowledge, such as auto-processed syntactic information.
• Previous methods of integrating syntactic information were inflexible, potentially degrading performance.

Purpose of the Study:

• To develop an advanced BioNER model that effectively leverages auto-processed syntactic information.
• To address the limitations of previous methods in flexibly incorporating syntactic data.
• To improve the accuracy and performance of BioNER systems in biomedical text analysis.

Main Methods:

• Proposed BIOKMNER, a BioNER model incorporating auto-processed syntactic information via key-value memory networks (KVMN).
• Utilized powerful encoders like BioBERT in conjunction with the KVMN approach.
• Evaluated the model on six diverse English biomedical datasets.

Main Results:

• BIOKMNER consistently outperformed the strong BioBERT baseline across all six datasets.
• Achieved state-of-the-art performance on four datasets: BC2GM, BC5CDR-chemical, NCBI-disease, and Species-800.
• Reported high F1 scores, including 85.29% on BC2GM and 94.22% on BC5CDR-chemical.

Conclusions:

• Auto-processed syntactic information is a valuable resource for enhancing BioNER.
• The proposed KVMN method effectively integrates syntactic information, leading to improved BioNER model performance.
• Experimental validation on benchmark datasets confirms the efficacy of the BIOKMNER model.